基于改进遗传算法与LS-SVM的变压器故障气体预测方法

最小二乘支持向量机(LS-SVM)能较好地解决小样本、非线性数据特征的多分类问题,适用于电力变压器油色谱故障气体预测,但参数c与σ2的选取对预测结果影响较大,有必要对其进行优化选择。笔者提出一种基于改进遗传算法(IGA)的参数寻优方法,并将其应用到变压器油中故障气体预测。IGA算法采用了编码机制,随机产生初始种群,可快速扩大搜索空间,稳定群体中个体多样性,有效提高全局搜索能力和收敛速度。最后进行了多组现场数据的实例分析,结果表明:基于IGA进行参数优化后的预测准确率明显优于传统LS-SVM预测结果。

Predicting Method for Dissolved Gas in Transformer Oil Based on Improved Genetic Algorithm and LS-SVM

LS-SVM(least square support vector machine) is applied to solve the multi-classification problems of small samples and non-linear data,c and σ2 are suitable for predicting dissolved gas in transformer oil.However,the selection of the parameters has clear impacts on the result of prediction,so it is necessary to optimize those parameters.In this paper,a new method to optimize those parameters based on IGA(improved genetic algorithm) is proposed and applied to predict dissolved gas in transformer oil.The IGA uses the encoding mechanism to randomly generates the initial population,rapidly expands the search space,stabilizes the diversity of the individuals in population,and effectively improves the global search ability and convergence speed.Case analyses of some sets of oil chromatogram data demonstrate that the prediction accuracy of the IGA-based LS-SVM is better than that of the conventional LS-SVM model.